{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2608.00915", "kind": "paper", "url": "https://arxiv.org/abs/2608.00915"}, {"id": "gh:binshuangli/uplift-bench", "kind": "repo", "url": "https://github.com/binshuangli/uplift-bench"}], "authors": ["Binshuang Li"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.00915", "source_url": "https://arxiv.org/abs/2608.00915"}, "categories": ["Language & Knowledge", "Software & AI Compute", "cs.LG"], "description": {"en": "A benchmark that evaluates uplift (conditional-average-treatment-effect) estimators across seven dataset families under a fixed, test-isolated, multi-objective protocol, with a public living leaderboard."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.00915", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.00915", "title": "UpliftBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2608.00915", "modality": null, "name": "UpliftBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_upliftbench_2d5414c6", "origin_source": "arxiv", "recognition_confidence": "0.75", "record_sha256": "d3cb68ae7a04ab674c1637cc118981840fa8f549c29738d202d0d25f678c6fcd", "relation": "introduces", "review_model": "claude-code-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T16:38:13.678431Z", "source_url": "https://arxiv.org/abs/2608.00915"}, "publisher": null, "released": "2026-08-02", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.00915", "source_url": "https://arxiv.org/abs/2608.00915"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-00915", "source": "claire_radar", "source_benchmark_id": "2608.00915", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Machine Generated", "applicationDomains": ["Software & AI Compute"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The living leaderboard and public repository support reuse, but the audience for causal-inference and uplift evaluation is specialized.", "score": 30}, "capabilities": [], "confidence": "Medium", "construction": "Original Synthetic", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T16:38:13.678431Z", "inputHash": "58d6f8ef4c6a177910f90439641a6d4ec6a68e5ab54d3cfc972aa136381e4850", "model": "claude-code-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "UpliftBench: Revealing Outcome-Regime and Objective Mismatch in Uplift Evaluation", "canonicalNameSource": "paper_title", "decisionReason": "It is a formally named benchmark with fixed protocols, versioned loaders, and a reproducible living leaderboard, all released publicly, which gives it a stable comparable scoring contract and a clear reuse path.", "evaluationEvidence": {"reuse": "UpliftBench releases versioned loaders, fixed protocols, result artifacts, and a reproducible living leaderboard", "scoring": "fixed protocols", "task": "UpliftBench evaluates 12 uplift estimators under an outer-test-isolated, multi-objective protocol across seven dataset families"}, "model": "claude-code-default", "reviewedAt": "2026-09-24T16:38:13.678431Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A benchmark that evaluates uplift (conditional-average-treatment-effect) estimators across seven dataset families under a fixed, test-isolated, multi-objective protocol, with a public living leaderboard.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "evaluation protocol evidence", "public artifact URL"], "snippet": "UpliftBench evaluates 12 uplift estimators under an outer-test-isolated, multi-objective protocol across seven dataset families; its two findings are identified where a reference objective exists -- F1 on the standard continuous benchmark (IHDP), F2 in a within-sample case study on Jobs."}, "familyId": "bmf_113036a53866", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_upliftbench_2d5414c6", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Software & Cloud"], "links": {"code": "https://github.com/binshuangli/uplift-bench", "data": null, "paper": "https://arxiv.org/abs/2608.00915", "pdf": "https://arxiv.org/pdf/2608.00915", "project": null, "report": "https://arxiv.org/abs/2608.00915"}, "metrics": [], "motivation": "Uplift modeling (conditional-average-treatment-effect estimation) drives personalized targeting, yet published uplift benchmarks frequently disagree on which estimator performs best; we show the disagreement is substantially about metrics, not models.", "name": "UpliftBench", "oneLine": "A benchmark that evaluates uplift (conditional-average-treatment-effect) estimators across seven dataset families under a fixed, test-isolated, multi-objective protocol, with a public living leaderboard.", "paperTitle": "UpliftBench: Revealing Outcome-Regime and Objective Mismatch in Uplift Evaluation", "primaryDomain": "Software & AI Compute", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.00915", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Runnable", "recognitionConfidence": 0.75, "relation": "introduces", "releasedAt": "2026-08-02", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Binshuang Li"], "categories": ["cs.LG"], "id": "2608.00915", "title": "UpliftBench: Revealing Outcome-Regime and Objective Mismatch in Uplift Evaluation", "type": "arxiv", "url": "https://arxiv.org/abs/2608.00915"}, "sourceUpdatedAt": "2026-08-02T00:51:44Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "It shows that disagreement among published uplift benchmarks comes largely from the choice of metric. It also gives the field versioned, reproducible protocols for comparing estimators."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
